A quick look at Facebook's HipHop Virtual Machine and the Hack language.
Size: 423.58 KB
Language: en
Added: May 06, 2014
Slides: 44 pages
Slide Content
HHVM + HACK
A quick introduction
HHVM
•Stands for HipHop Virtual Machine
•Executes scripts written in PHP (5.4) and Hack
•Drop-in replacement for php-fpm
•Use together with FastCGI-enabled webserver
(like Apache or nginx)
Timeline
•2008: Facebook begins work on HipHop, a PHP to C++ compiler
•2010: Facebook opensources the compiler (HPHPc) and dev interpreter/
debugger (HPHPi and HPHPd)
•2010: Facebook begins work on HHVM
•2013: HPHPc deprecated, replaced by HHVM in Facebook production servers
•2013: HHVM 2.3.0 adds FastCGI support
•2014: HHVM 3.0.0 adds Hack support
Installation
•Ubuntu
add-apt-repository -y ppa:mapnik/boost!
wget -O - http://dl.hhvm.com/conf/hhvm.gpg.key | sudo apt-key add -!
echo "deb http://dl.hhvm.com/ubuntu precise main" > /etc/apt/sources.list.d/hhvm.list!
apt-get update!
apt-get install hhvm!
•Note: there is no more hhvm-fastcgi package. For the latest
code, you can use hhvm-nightly
•https://github.com/facebook/hhvm/wiki/Prebuilt-Packages-on-
Ubuntu-12.04
Installation
•OSX (using Homebrew)
brew tap homebrew/dupes!
brew tap homebrew/versions!
brew tap mcuadros/homebrew-hhvm!
brew install hhvm --HEAD !
•https://github.com/facebook/hhvm/wiki/Building-and-
installing-HHVM-on-OSX-10.9
•Unfortunately, the typechecker (hh_server / hh_client)
does not build on OSX
Installation
•Windows
•… just use Vagrant and Virtualbox
Usage
•Using from the command line
hhvm foobar.php
•Or you can make use of this:
sudo /usr/bin/update-alternatives --install \
/usr/bin/php php /usr/bin/hhvm 60
•Now you can run
php foobar.php
Usage
•nginx: just like your regular php-fpm
location ~ \.(hh|php)$ {!
root /path/to/your/root!
fastcgi_pass 127.0.0.1:9000;!
fastcgi_index index.php;!
fastcgi_param SCRIPT_FILENAME $document_root$fastcgi_script_name;!
include /etc/nginx/fastcgi_params;!
}
Using with Composer
•Add to .bashrc or equivalent:
alias composer="hhvm -v ResourceLimit.SocketDefaultTimeout=30
-v Http.SlowQueryThreshold=30000 /usr/local/bin/composer"
•Up to 5x faster
Extensions
•HHVM ships with several common extensions, listed here:
https://github.com/facebook/hhvm/wiki/Extensions
•Usually, to add an extension, you write it in pure PHP (e.g. the Redis
extension) and place it alongside the source code, to be compiled
together with HHVM
•You may also use HNI (HHVM-Native Interface) for hybrid PHP/C++
implementations
•Can be fashioned as an externally buildable DSO, e.g. mongofill
Performance
•Up to 2x faster compared to PHP 5.5
(Sample benchmark performed by Christian Stocker with Symfony 2.4)
http://blog.liip.ch/archive/2013/10/29/hhvm-and-symfony2.html
Caveats
•PHP 5.5 generator syntax incompatibility (#1627, #1787, #1871)
•Missing PHP 5.4 Closure::bind and Closure::bindTo (#1203)
•Cannot set multiple cookies with same name but different path (#2494, #2526)
•func_get_args() returns arguments by references instead of values (#1027) (wontfix)
•array_key_exists(), reset(), end(), etc. don't work with ArrayAccess (#1221) (wontfix)
•missing fastcgi_finish_request() equivalent (#1230)
•https://github.com/facebook/hhvm/issues?labels=php5+incompatibility&state=open
Hack
•Basically a superset of PHP, the most important addition being static typing
•Hack files should begin with <?hh instead of <?php
•Can be used in 3 modes: "strict", "partial" (default) and "decl", e.g. <?hh //
strict
•"// UNSAFE" annotation used to disable type checking for a particular block
•Decl mode basically disables type checking for the entire file, enables strict
mode to call into legacy code
•Type annotations are thrown away at runtime
Type annotations
•This allows validation by the Hack typechecker before runtime
•Paired with IDE support, can be helpful when refactoring code
•Scalar type hinting is now possible
class MyClass {!
const int MyConst = 0;!
private string $x = '';!
!
public function increment(int $x): int {!
$y = $x + 1;!
return $y;!
}!
}
Supported types
•XHP elements :x:frag, :x:base, :div, :xhp
•Generics: Foo<T>
•Closures: (function(<type1>, <type2>, ...): return_type)
e.g. (function(string, int): string)
•Resources: resource
•Continuation (generator): Continuation<T>
•Awaitable (async): Awaitable<T>
•Same type: this
Supported types
•Note that:
•Untyped arrays are not allowed in strict mode
•Do not annotate return values for __construct()
(typechecker will complain)
•"this" only checks for same type, not same instance
Generics
class Box<T> {!
protected T $data;!
!
public function __construct(T $data) {!
$this->data = $data;!
}!
public function setData(T $data): void {!
$this->data = $data;!
}!
public function getData(): T {!
return $this->data;!
}!
}!
!
function swap<T>(Box<T> $a, Box<T> $b): T {!
$temp = $a->getData();!
$a->setData($b->getData());!
$b->setData($temp);!
return $temp;!
}!
•For each instance, once a type is associated with T, it’s fixed
Nullable types
class NullableBox<T> {!
protected ?T $data;!
!
public function __construct(?T $data) {!
$this->data = $data;!
}!
!
public function getData(): ?T {!
return $this->data;!
}!
}!
!
$box = new NullableBox(null);!
!
!
function check_not_null(?int $x): int {!
return $x === null ? -1 : $x;!
}!
Soft types
•Apparently not documented
•Emits warnings rather than fatal errors for
invalid types
!
!
class Calculator {!
public function add(@int $a, @int $b): @string {}!
}!
!
$calc = new Calculator();!
$calc->add("1", "2");
Typed arrays
•Untyped arrays are only allowed in partial or decl mode
•In strict mode you must explicitly type the values:
array<Foo>!
•Or both keys and values:
array<string, ?string>!
•You can simulate PHP behaviour, but this defeats the
purpose:
array<mixed, mixed>
Type aliasing
•Redefine an existing type name
type MyInteger = int;!
•“Opaque” types disallow other files from accessing the
underlying implementation (e.g. concatenation for strings)
// File1.php!
newtype MyString = string;!
!
// File2.php!
require_once "File1.php";!
!
function modifyString(MyString $s) : MyString {!
return $s . "1"; // Hack type checker will throw an error!
}
Tuples and shapes
•Tuples are basically immutable typed arrays
public function baz(): (string, int) {!
return tuple("Hello", 3);!
}!
•Shapes are kinda like structs
type MyShape = shape('id1' => string, 'id2' => int);!
function foo(MyShape $x): void {}
Collections
•Vector, Map, Set, Pair
•Immutable versions: ImmVector, ImmMap, ImmSet
•Can be initialised using literal syntax
$vec = Vector {'foo', 'foo', 'bar'}; // integer keys!
$map = Map {42 => 'foo', 73 => 'bar', 144 => 'baz'}; // ordered dictionary!
$set = Set {'php', 'hack', 'hhvm'}; // unordered unique values!
$par = Pair {'guid', 'ABC123'}; // only two pieces of data !
•Type annotation
function getTags(): Set<string> {!
return Set { "php", "hack", "hhvm" };!
}
Collections
•Some method signatures for Vector…
public function __construct(?Traversable<Tv> $it)!
public function add(Tv $value): Vector<Tv>!
public function addAll(?Traversable<Tv> $it): Vector<Tv>!
public function at(int $k): Tv!
public function clear(void): Vector<Tv>!
public function containsKey(int $k): bool!
public function count(void): int!
public function filter((function(Tv): bool) $callback): Vector<Tv>!
public function filterWithKey((function(int, Tv): bool) $callback): Vector<Tv>!
public function fromArray(array $arr): Vector<Tv>!
public function fromItems(?Traversable<Tv> $items): Vector<Tv>!
public function get(int $k): ?Tv!
public function getIterator(void): KeyedIterator<int, Tv> !
public function isEmpty(void): bool!
public function items(void): Iterable<Tv>!
public function keys(void): Vector<int>!
public function reverse(void): void!
public function set(int $k, Tv $v): Vector<Tv>!
public function setAll(?KeyedTraversable<int, Tv> $it): Vector<Tv>!
public function __toString(void): string!
public function toValuesArray(void): array!
public function zip<Tu>(Traversable<Tu> $iterable): Vector<Pair<Tv, Tu>>
In other words, Hack really, really wants you
to stop using simple arrays (or even
ArrayObjects) as a “catch-all” data container
Lambdas
•Use the lambda arrow ==>
(Because -> and => are already taken):
function concat(): (function(string, string): string) {!
return ($x, $y) ==> {!
return $y . $x;!
};!
}!
•Implicitly capture variables from outer scope
Can be an expression:
function foo(): (function(string): string) {!
$x = 'bar';!
return $y ==> $y . $x;!
}
Lambdas
•Type hints are not checked at the moment
•Doesn't support capturing variables by reference or
returning by reference yet
•Lambda arrow operator is right associative and can be
chained:
!
$f = $x ==> $y ==> $x + $y;!
$g = $f(7);!
echo $g(4); // 11
Variadic functions
function sum(...): int {!
$s = 0;!
foreach (func_get_args() as $e) {!
$s += $e;!
}!
return $s;!
}!
•Cannot be type annotated, unfortunately
Override attribute
•Signify that the parent method must exist
class Hello {!
public function render(): void {!
echo 'hello';!
}!
}!
!
class HelloWorld extends Hello!
{!
<<Override>> public function render(): void {!
parent::render();!
echo ' world';!
}!
}
Constructor arg promotion
class Person {!
private string $name;!
private int $age;!
!
public function __construct(string $name, int $age) {!
$this->name = $name;!
$this->age = $age;!
}!
}!
•can be written as
!
class Person {!
public function __construct(private string $name, private int $age) {}!
}
Callbacks
•This will not work in strict mode:
$func = 'myFunc';!
$x = $func(4);!
•fun() will return a callable, which allows the typechecker to properly validate:
$func = fun('myFunc');!
$x = $func(4);!
•Similar dubiously-named functions exist for calling static/instance methods:
// $bar = array_map(array('Foo', ‘doStuff'), [1, 2, 3] );!
$bar = array_map(class_meth('Foo', 'doStuff'), [1, 2, 3]);!
!
$foo = new Foo();!
$bar = array_map(instance_meth($foo, 'doStuff'), [1, 2, 3]);!
!
$func = meth_caller('Foo', 'doStuff');!
$foo = new Foo();!
$x = $func($foo);
Async and await
async function gen_foo(int $a): Awaitable<?Foo> {!
if ($a === 0) {!
return null;!
}!
$bar = await gen_bar($a);!
if ($bar !== null) {!
return $bar->getFoo();!
}!
return null;!
}!
!
async function gen_bar(int $a): Awaitable<?Bar> {!
if ($a === 0) {!
return null;!
}!
return new Bar();!
}!
!
gen_foo(4);
Running the type checker
•Run hh_client (it will launch hh_server for you) or
hh_server --check .
•Recursively looks for directories that contain
a .hhconfig file (it won’t run if this is missing)
•Optionally return output in json format
•Note: does not follow symlinks
Unsupported features
•Still runs in HHVM, but the type checker does not like the following:
•goto, if: … endif; etc
•AND, OR, XOR
•References (&)
•Error suppression (@)
•break N, continue N
•eval()
•globals
Unsupported features
•list(, $b) = array(3, 4); should be list($_, $b) = array(3, 4);
•Function calls are case-sensitive
•Strict mode does not support top-level code except for require() and
require_once(); i.e. everything else must be in a class or function
•Does not support "variable variables" (e.g. $$a) or extract()
•Cannot call parent::staticMethod()
•Cannot declare a class method with a name that collides with class name
•Cannot pass primitives by reference
Drawbacks
•Still new, not yet 100% compatibility (but catching up fast)
•Your favourite extension may not be available
•Limited IDE support at the moment (weekend project idea?)
•Hack-exclusive libraries? Package repos? Community
fragmentation?
•Clashes with PHP 5.5 and future 5.6 syntax
(generators and variadic functions)
•Generator functions are of the type
“Continuation” in HHVM (since they were
implemented earlier)
•The following is valid in PHP 5.5 but not in HHVM:
$data = (yield $value); // delete the brackets for HHVM!
$data = yield; // HHVM needs expression after yield !
•HHVM Continuations need to be “primed” unlike
PHP 5.5 Generators
$generator = construct_generator();!
$generator->current(); // must call next() first in HHVM!
•Proposed PHP 5.6 syntax for variadic
functions:
function fn($arg, ...$args) {} !
function fn($arg, &...$args) {}!
function fn($arg, callable ...$args) {} !
•Type-hinting and arguments by value not
supported in HHVM / Hack
function fn(...): void {}
Benefits
•Type safety!!! (Always use the type checker)
•Syntactic sugar from other languages
•Discard “bad” parts of PHP
•Active community (e.g. New Relic has released a
prototype HHVM extension, Heroku just announced
HHVM support)